GEO · 6 MIN
How AI assistants recommend agencies and what you can do about it
What OpenAI, Anthropic, and Perplexity document about how assistants use the web, and a practical plan for making your agency easier to find and describe.
AI assistants recommend agencies by running searches or retrieving pages, then summarizing what they find. The vendors document their crawlers and robots.txt controls, but none publish a ranking formula. So the reliable approach is to make your site crawlable, say plainly what you do and for whom, and test what assistants say in a fresh session.
- OpenAI and Anthropic document separate bots for training, search, and user-requested fetching. Perplexity documents a search bot and a user-requested fetcher, and says its search bot is not used to crawl content for AI foundation models.
- Blocking a search bot can keep your pages out of answers, while blocking a training bot is a separate choice.
- No vendor publishes how recommendations are ranked, so treat any "secret factor" claim as unverified.
- Specific, quotable pages about your services, location, and process are easier for an assistant to summarize accurately.
- You can measure your position by asking buyer questions cold and recording the results.
- Nactore is an AI-native technology and growth company that builds crawlable sites and runs GEO and SEO content as one program.
How do assistants find agencies at all?
There are two sources. The model's built-in knowledge comes from training and is frozen at a point in time. Many assistants also search or fetch web pages live to answer a question, then cite what they found. For a query like "AI agency in Dubai," a live-search assistant looks at pages available to its search tools, so your web presence matters directly.
The vendors describe this in their own words. OpenAI's crawler overview lists separate user agents for different purposes. Anthropic's crawler article and Perplexity's crawler page do the same. Read them, because the details change.
What do the vendors actually document?
| Vendor | Bots named in documentation | Purpose as described by the vendor |
|---|---|---|
| OpenAI | GPTBot, OAI-SearchBot, ChatGPT-User (plus OAI-AdsBot for ad landing pages) | Training, surfacing sites in ChatGPT search, and user-initiated page visits |
| Anthropic | ClaudeBot, Claude-User, Claude-SearchBot | Training collection, fetching pages when a user asks, and improving search results |
| Perplexity | PerplexityBot, Perplexity-User | Surfacing and linking sites in Perplexity search (not used to crawl content for AI foundation models), and visiting pages to answer a user's question |
Two things follow. First, the vendors separate search from training, so you can make different robots.txt choices for each. Note that OpenAI says robots.txt rules may not apply to user-initiated ChatGPT-User visits, and Perplexity says Perplexity-User generally ignores robots.txt, because a person asked for the page. Second, the documentation says nothing about which agencies are recommended. Anyone describing exact ranking factors is guessing, and you should discount it.
Check the current wording on each vendor's page before you edit robots.txt, since names and behavior get updated.
Why might an assistant not mention your agency?
The usual causes are mundane, and most are fixable.
- It cannot reach your site. A firewall, a country block, or a blanket robots.txt rule stops the bots. Check your server logs for the documented user agents.
- The page does not say it. If your services page uses vague slogans, an assistant has nothing concrete to quote. "Digital solutions for modern brands" says nothing. "Shopify builds and SEO for UK retailers" does.
- The key facts live in images or scripts. Text rendered only by JavaScript or embedded in graphics may not be read.
- Other sources describe you differently. Directories, review sites, and your own social profiles feed the same picture.
- You are new or small. Assistants often have less to draw on for young brands, which is a reason to publish evidence and be listed in places buyers and journalists use.
What should an agency page contain to be quotable?
An assistant summarizing a service provider needs a handful of facts. Put them in plain text, near the top of the page.
- What you do. One sentence naming the services and the buyer type.
- Where you work. Countries, cities, and time zones you serve.
- How you work. Your process in numbered steps, with what the client provides and receives.
- Who runs it. Real names, roles, and a short bio for the people leading the work.
- Proof. Real clients you may name, published results with their source, or honest statements of what you have built. Never invent either.
- How to start. A booking link and what happens on the first call.
Back these with question-style headings, since buyers type questions and assistants lift answers under matching headings. Our post on why AI tools cite some brands goes deeper on the evidence side.
What can you do outside your own site?
Assistants form a picture from many pages. Tend the places buyers and publishers already trust.
- Profiles and directories. Keep company name, services, and location identical everywhere.
- Mentions in real publications. Write useful pieces or contribute quotes for trade outlets that cover your niche.
- Communities. Answer questions honestly where your buyers ask them, and disclose your affiliation.
- Reviews. Ask real clients for reviews on the platforms your buyers read. Do not buy or fake them.
None of these are shortcuts. They do give an assistant more consistent evidence to draw on. For a longer playbook, read how to get mentioned in ChatGPT.
How do you measure whether it works?
Test the way a buyer would. Use a fresh session, with no history, and ask the questions your clients ask.
- Write 10 to 15 buyer questions. Examples include "who builds AI automation for law firms" or "how to choose a software agency in Singapore."
- Run each in the assistants your buyers use. Record the date, the exact wording, whether you are named, and which pages are cited.
- Repeat monthly. Answers vary between runs, so look at patterns across many runs, not one result.
- Check referral traffic. Look for visits from assistant domains in your analytics.
We describe the method in how to measure AI search visibility.
Do not infer what an assistant does from web searches about it. Ask the assistant the actual question and record what it returns, because that is the only evidence that counts.
Frequently asked questions
Can I pay to be recommended by an AI assistant?
The vendors document their crawlers and search features, and none describe a way to buy a place in an assistant's recommendation. Be cautious of anyone who says they can sell one.
Should I block training bots?
That is a business decision. OpenAI and Anthropic document training bots separately from their search bots, so you can allow search access while declining training. Perplexity says its search bot is not used for foundation model training. Review current documentation before changing robots.txt.
Does schema markup help assistants?
Structured data helps Google understand pages, and clean, accurate markup does no harm elsewhere. The assistant vendors do not document it as a ranking input, so treat it as hygiene, not a lever.
How long does it take to show up?
It depends on crawling, your existing authority, and the question asked, so no honest number exists. Measure monthly and look for direction.
Make it easy to be right about you
Assistants cannot recommend what they cannot read or describe. Fix access, state your services plainly, build consistent evidence elsewhere, and measure with real questions. Nactore treats this as one program with SEO.
Want this built for your team? Book a free 30-minute call.
Want to apply this to your business?
Book a free 30-minute call. We will tell you what we would do first.